From patchwork Fri Dec 18 03:43:28 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Patrick McGehearty X-Patchwork-Id: 1418032 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=gcc.gnu.org (client-ip=2620:52:3:1:0:246e:9693:128c; helo=sourceware.org; envelope-from=gcc-patches-bounces@gcc.gnu.org; receiver=) Authentication-Results: ozlabs.org; dmarc=pass (p=none dis=none) header.from=gcc.gnu.org Authentication-Results: ozlabs.org; dkim=pass (1024-bit key; unprotected) header.d=gcc.gnu.org header.i=@gcc.gnu.org header.a=rsa-sha256 header.s=default header.b=xLB1eqZm; dkim-atps=neutral Received: from sourceware.org (server2.sourceware.org [IPv6:2620:52:3:1:0:246e:9693:128c]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 4CxvqP2PpRz9sVq for ; Fri, 18 Dec 2020 14:43:58 +1100 (AEDT) Received: from server2.sourceware.org (localhost [IPv6:::1]) by sourceware.org (Postfix) with ESMTP id 8B86D389851D; Fri, 18 Dec 2020 03:43:49 +0000 (GMT) DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 8B86D389851D DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org; s=default; t=1608263029; bh=wSkdSdoby36NNUA462xwQn0iWEGSZqXjjaMR5350yCQ=; h=To:Subject:Date:List-Id:List-Unsubscribe:List-Archive:List-Post: List-Help:List-Subscribe:From:Reply-To:From; b=xLB1eqZmAXOR80tfh5x9GEaOAJNOM3ChL6o55UiBfdCXtsNITMIFP0/LXK6rWL8J4 SPJjkKgvxdk9BL425b83FJ5wcD73+E55FXb5xYbWnxr37iQvBAoWzifmZaD3m+2JPc lKCifjM3rxUE/Ug76uQo6WO59AkS92h+zJ5afYh4= X-Original-To: gcc-patches@gcc.gnu.org Delivered-To: gcc-patches@gcc.gnu.org Received: from userp2130.oracle.com (userp2130.oracle.com [156.151.31.86]) by sourceware.org (Postfix) with ESMTPS id 2C05F389851D for ; Fri, 18 Dec 2020 03:43:43 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org 2C05F389851D Received: from pps.filterd (userp2130.oracle.com [127.0.0.1]) by userp2130.oracle.com (8.16.0.42/8.16.0.42) with SMTP id 0BI3evKQ103678 for ; Fri, 18 Dec 2020 03:43:39 GMT Received: from aserp3020.oracle.com (aserp3020.oracle.com [141.146.126.70]) by userp2130.oracle.com with ESMTP id 35cn9rrgks-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=FAIL) for ; Fri, 18 Dec 2020 03:43:38 +0000 Received: from pps.filterd (aserp3020.oracle.com [127.0.0.1]) by aserp3020.oracle.com (8.16.0.42/8.16.0.42) with SMTP id 0BI3ZJUo139369 for ; Fri, 18 Dec 2020 03:43:37 GMT Received: from userv0121.oracle.com (userv0121.oracle.com [156.151.31.72]) by aserp3020.oracle.com with ESMTP id 35e6eu6enx-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK) for ; Fri, 18 Dec 2020 03:43:37 +0000 Received: from abhmp0016.oracle.com (abhmp0016.oracle.com [141.146.116.22]) by userv0121.oracle.com (8.14.4/8.13.8) with ESMTP id 0BI3hXBF019792 for ; Fri, 18 Dec 2020 03:43:36 GMT Received: from arm34vm.us.oracle.com (/10.211.14.34) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Thu, 17 Dec 2020 19:43:33 -0800 To: gcc-patches@gcc.gnu.org Subject: [PATCH v6] Practical improvement to libgcc complex divide Date: Fri, 18 Dec 2020 03:43:28 +0000 Message-Id: <1608263008-2556-1-git-send-email-patrick.mcgehearty@oracle.com> X-Mailer: git-send-email 1.8.3.1 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9838 signatures=668683 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 adultscore=0 mlxscore=0 phishscore=0 bulkscore=0 suspectscore=0 malwarescore=0 mlxlogscore=999 spamscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2009150000 definitions=main-2012180024 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9838 signatures=668683 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 adultscore=0 mlxlogscore=999 impostorscore=0 lowpriorityscore=0 clxscore=1015 spamscore=0 malwarescore=0 priorityscore=1501 phishscore=0 mlxscore=0 bulkscore=0 suspectscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2009150000 definitions=main-2012180024 X-Spam-Status: No, score=-11.5 required=5.0 tests=BAYES_00, DKIMWL_WL_HIGH, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, GIT_PATCH_0, KAM_ASCII_DIVIDERS, KAM_SHORT, RCVD_IN_MSPIKE_H2, SPF_HELO_PASS, SPF_PASS, TXREP, UNPARSEABLE_RELAY autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-Patchwork-Original-From: Patrick McGehearty via Gcc-patches From: Patrick McGehearty Reply-To: Patrick McGehearty Errors-To: gcc-patches-bounces@gcc.gnu.org Sender: "Gcc-patches" Changes in this version from Version 5: Commit message change: Information for Changelogs put in proper format. Thanks go to Jakub Jelinek for enlightening me how to do that and providing an excellent example. c-cppbuiltin.c - Fixed various formatting issues regarding misplaced or missing spaces. Replaced strncpy with memcpy. Replaced all uses of strlen (name) with name_len in calls to alloca. Did not replace alloca with XALLOCAVEC. There are no other uses of XALLOCAVEC in this file, although XALLOCAVEC was used in c-ada-spec.c and c-common.c. Following the principle of least change consistent coding practice within a single source file I am hesitant to make the change. If gcc maintainers desire to generally replace all uses of alloca with XALLOCVEC at this time for this file, I'm willing to make the change. No changes to the other modified files from Version 5. Summary of Purpose This patch to libgcc/libgcc2.c __divdc3 provides an opportunity to gain important improvements to the quality of answers for the default complex divide routine (half, float, double, extended, long double precisions) when dealing with very large or very small exponents. The current code correctly implements Smith's method (1962) [2] further modified by c99's requirements for dealing with NaN (not a number) results. When working with input values where the exponents are greater than *_MAX_EXP/2 or less than -(*_MAX_EXP)/2, results are substantially different from the answers provided by quad precision more than 1% of the time. This error rate may be unacceptable for many applications that cannot a priori restrict their computations to the safe range. The proposed method reduces the frequency of "substantially different" answers by more than 99% for double precision at a modest cost of performance. Differences between current gcc methods and the new method will be described. Then accuracy and performance differences will be discussed. Background This project started with an investigation related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59714. Study of Beebe[1] provided an overview of past and recent practice for computing complex divide. The current glibc implementation is based on Robert Smith's algorithm [2] from 1962. A google search found the paper by Baudin and Smith [3] (same Robert Smith) published in 2012. Elen Kalda's proposed patch [4] is based on that paper. I developed two sets of test data by randomly distributing values over a restricted range and the full range of input values. The current complex divide handled the restricted range well enough, but failed on the full range more than 1% of the time. Baudin and Smith's primary test for "ratio" equals zero reduced the cases with 16 or more error bits by a factor of 5, but still left too many flawed answers. Adding debug print out to cases with substantial errors allowed me to see the intermediate calculations for test values that failed. I noted that for many of the failures, "ratio" was a subnormal. Changing the "ratio" test from check for zero to check for subnormal reduced the 16 bit error rate by another factor of 12. This single modified test provides the greatest benefit for the least cost, but the percentage of cases with greater than 16 bit errors (double precision data) is still greater than 0.027% (2.7 in 10,000). Continued examination of remaining errors and their intermediate computations led to the various tests of input value tests and scaling to avoid under/overflow. The current patch does not handle some of the rare and most extreme combinations of input values, but the random test data is only showing 1 case in 10 million that has an error of greater than 12 bits. That case has 18 bits of error and is due to subtraction cancellation. These results are significantly better than the results reported by Baudin and Smith. Support for half, float, double, extended, and long double precision is included as all are handled with suitable preprocessor symbols in a single source routine. Since half precision is computed with float precision as per current libgcc practice, the enhanced algorithm provides no benefit for half precision and would cost performance. Further investigation showed changing the half precision algorithm to use the simple formula (real=a*c+b*d imag=b*c-a*d) caused no loss of precision and modest improvement in performance. The existing constants for each precision: float: FLT_MAX, FLT_MIN; double: DBL_MAX, DBL_MIN; extended and/or long double: LDBL_MAX, LDBL_MIN are used for avoiding the more common overflow/underflow cases. This use is made generic by defining appropriate __LIBGCC2_* macros in c-cppbuiltin.c. Tests are added for when both parts of the denominator have exponents small enough to allow shifting any subnormal values to normal values all input values could be scaled up without risking overflow. That gained a clear improvement in accuracy. Similarly, when either numerator was subnormal and the other numerator and both denominator values were not too large, scaling could be used to reduce risk of computing with subnormals. The test and scaling values used all fit within the allowed exponent range for each precision required by the C standard. Float precision has more difficulty with getting correct answers than double precision. When hardware for double precision floating point operations is available, float precision is now handled in double precision intermediate calculations with the simple algorithm the same as the half-precision method of using float precision for intermediate calculations. Using the higher precision yields exact results for all tested input values (64-bit double, 32-bit float) with the only performance cost being the requirement to convert the four input values from float to double. If double precision hardware is not available, then float complex divide will use the same improved algorithm as the other precisions with similar change in performance. Further Improvement The most common remaining substantial errors are due to accuracy loss when subtracting nearly equal values. This patch makes no attempt to improve that situation. NOTATION For all of the following, the notation is: Input complex values: a+bi (a= real part, b= imaginary part) c+di Output complex value: e+fi = (a+bi)/(c+di) For the result tables: current = current method (SMITH) b1div = method proposed by Elen Kalda b2div = alternate method considered by Elen Kalda new = new method proposed by this patch DESCRIPTIONS of different complex divide methods: NAIVE COMPUTATION (-fcx-limited-range): e = (a*c + b*d)/(c*c + d*d) f = (b*c - a*d)/(c*c + d*d) Note that c*c and d*d will overflow or underflow if either c or d is outside the range 2^-538 to 2^512. This method is available in gcc when the switch -fcx-limited-range is used. That switch is also enabled by -ffast-math. Only one who has a clear understanding of the maximum range of all intermediate values generated by an application should consider using this switch. SMITH's METHOD (current libgcc): if(fabs(c) RBIG) || (FABS (a) > RBIG) || (FABS (b) > RBIG) ) { a = a * 0.5; b = b * 0.5; c = c * 0.5; d = d * 0.5; } /* minimize overflow/underflow issues when c and d are small */ else if (FABS (d) < RMIN2) { a = a * RMINSCAL; b = b * RMINSCAL; c = c * RMINSCAL; d = d * RMINSCAL; } else { if(((FABS (a) < RMIN) && (FABS (b) < RMAX2) && (FABS (d) < RMAX2)) || ((FABS (b) < RMIN) && (FABS (a) < RMAX2) && (FABS (d) < RMAX2))) { a = a * RMINSCAL; b = b * RMINSCAL; c = c * RMINSCAL; d = d * RMINSCAL; } } r = c/d; denom = (c*r) + d; if( r > RMIN ) { e = (a*r + b) / denom ; f = (b*r - a) / denom } else { e = (c * (a/d) + b) / denom; f = (c * (b/d) - a) / denom; } } [ only presenting the fabs(c) < fabs(d) case here, full code in patch. ] Before any computation of the answer, the code checks for any input values near maximum to allow down scaling to avoid overflow. These scalings almost never harm the accuracy since they are by 2. Values that are over RBIG are relatively rare but it is easy to test for them and allow aviodance of overflows. Testing for RMIN2 reveals when both c and d are less than [FLT|DBL]_EPSILON. By scaling all values by 1/EPSILON, the code converts subnormals to normals, avoids loss of accuracy and underflows in intermediate computations that otherwise might occur. If scaling a and b by 1/EPSILON causes either to overflow, then the computation will overflow whatever method is used. Finally, we test for either a or b being subnormal (RMIN) and if so, for the other three values being small enough to allow scaling. We only need to test a single denominator value since we have already determined which of c and d is larger. Next, r (the ratio of c to d) is checked for being near zero. Baudin and Smith checked r for zero. This code improves that approach by checking for values less than DBL_MIN (subnormal) covers roughly 12 times as many cases and substantially improves overall accuracy. If r is too small, then when it is used in a multiplication, there is a high chance that the result will underflow to zero, losing significant accuracy. That underflow is avoided by reordering the computation. When r is subnormal, the code replaces a*r (= a*(c/d)) with ((a/d)*c) which is mathematically the same but avoids the unnecessary underflow. TEST Data Two sets of data are presented to test these methods. Both sets contain 10 million pairs of complex values. The exponents and mantissas are generated using multiple calls to random() and then combining the results. Only values which give results to complex divide that are representable in the appropriate precision after being computed in quad precision are used. The first data set is labeled "moderate exponents". The exponent range is limited to -DBL_MAX_EXP/2 to DBL_MAX_EXP/2 for Double Precision (use FLT_MAX_EXP or LDBL_MAX_EXP for the appropriate precisions. The second data set is labeled "full exponents". The exponent range for these cases is the full exponent range including subnormals for a given precision. ACCURACY Test results: Note: The following accuracy tests are based on IEEE-754 arithmetic. Note: All results reporteed are based on use of fused multiply-add. If fused multiply-add is not used, the error rate increases, giving more 1 and 2 bit errors for both current and new complex divide. Differences between using fused multiply and not using it that are greater than 2 bits are less than 1 in a million. The complex divide methods are evaluated by determining the percentage of values that exceed differences in low order bits. If a "2 bit" test results show 1%, that would mean that 1% of 10,000,000 values (100,000) have either a real or imaginary part that differs from the quad precision result by more than the last 2 bits. Results are reported for differences greater than or equal to 1 bit, 2 bits, 8 bits, 16 bits, 24 bits, and 52 bits for double precision. Even when the patch avoids overflows and underflows, some input values are expected to have errors due to the potential for catastrophic roundoff from floating point subtraction. For example, when b*c and a*d are nearly equal, the result of subtraction may lose several places of accuracy. This patch does not attempt to detect or minimize this type of error, but neither does it increase them. I only show the results for Elen Kalda's method (with both 1 and 2 divides) and the new method for only 1 divide in the double precision table. In the following charts, lower values are better. current - current complex divide in libgcc b1div - Elen Kalda's method from Baudin & Smith with one divide b2div - Elen Kalda's method from Baudin & Smith with two divides new - This patch which uses 2 divides =================================================== Errors Moderate Dataset gtr eq current b1div b2div new ====== ======== ======== ======== ======== 1 bit 0.24707% 0.92986% 0.24707% 0.24707% 2 bits 0.01762% 0.01770% 0.01762% 0.01762% 8 bits 0.00026% 0.00026% 0.00026% 0.00026% 16 bits 0.00000% 0.00000% 0.00000% 0.00000% 24 bits 0% 0% 0% 0% 52 bits 0% 0% 0% 0% =================================================== Table 1: Errors with Moderate Dataset (Double Precision) Note in Table 1 that both the old and new methods give identical error rates for data with moderate exponents. Errors exceeding 16 bits are exceedingly rare. There are substantial increases in the 1 bit error rates for b1div (the 1 divide/2 multiplys method) as compared to b2div (the 2 divides method). These differences are minimal for 2 bits and larger error measurements. =================================================== Errors Full Dataset gtr eq current b1div b2div new ====== ======== ======== ======== ======== 1 bit 2.05% 1.23842% 0.67130% 0.16664% 2 bits 1.88% 0.51615% 0.50354% 0.00900% 8 bits 1.77% 0.42856% 0.42168% 0.00011% 16 bits 1.63% 0.33840% 0.32879% 0.00001% 24 bits 1.51% 0.25583% 0.24405% 0.00000% 52 bits 1.13% 0.01886% 0.00350% 0.00000% =================================================== Table 2: Errors with Full Dataset (Double Precision) Table 2 shows significant differences in error rates. First, the difference between b1div and b2div show a significantly higher error rate for the b1div method both for single bit errros and well beyond. Even for 52 bits, we see the b1div method gets completely wrong answers more than 5 times as often as b2div. To retain comparable accuracy with current complex divide results for small exponents and due to the increase in errors for large exponents, I choose to use the more accurate method of two divides. The current method has more 1.6% of cases where it is getting results where the low 24 bits of the mantissa differ from the correct answer. More than 1.1% of cases where the answer is completely wrong. The new method shows less than one case in 10,000 with greater than two bits of error and only one case in 10 million with greater than 16 bits of errors. The new patch reduces 8 bit errors by a factor of 16,000 and virtually eliminates completely wrong answers. As noted above, for architectures with double precision hardware, the new method uses that hardware for the intermediate calculations before returning the result in float precision. Testing of the new patch has shown zero errors found as seen in Tables 3 and 4. Correctness for float ============================= Errors Moderate Dataset gtr eq current new ====== ======== ======== 1 bit 28.68070% 0% 2 bits 0.64386% 0% 8 bits 0.00401% 0% 16 bits 0.00001% 0% 24 bits 0% 0% ============================= Table 3: Errors with Moderate Dataset (float) ============================= Errors Full Dataset gtr eq current new ====== ======== ======== 1 bit 19.98% 0% 2 bits 3.20% 0% 8 bits 1.97% 0% 16 bits 1.08% 0% 24 bits 0.55% 0% ============================= Table 4: Errors with Full Dataset (float) As before, the current method shows an troubling rate of extreme errors. There very mincr changes in accuracy for half-precision since the code changes from Smith's method to the simple method. 5 out of 1 million test cases show correct answers instead of 1 or 2 bit errors. libgcc computes half-precision functions in float precision allowing the existing methods to avoid overflow/underflow issues for the allowed range of exponents for half-precision. Extended precision (using x87 80-bit format on x86) and Long double (using IEEE-754 128-bit on x86 and aarch64) both have 15-bit exponents as compared to 11-bit exponents in double precision. We note that the C standard also allows Long Double to be implemented in the equivalent range of Double. The RMIN2 and RMINSCAL constants are selected to work within the Double range as well as with extended and 128-bit ranges. We will limit our performance and accurancy discussions to the 80-bit and 128-bit formats as seen on x86 here. The extended and long double precision investigations were more limited. Aarch64 does not support extended precision but does support the software implementation of 128-bit long double precision. For x86, long double defaults to the 80-bit precision but using the -mlong-double-128 flag switches to using the software implementation of 128-bit precision. Both 80-bit and 128-bit precisions have the same exponent range, with the 128-bit precision has extended mantissas. Since this change is only aimed at avoiding underflow/overflow for extreme exponents, I studied the extended precision results on x86 for 100,000 values. The limited exponent dataset showed no differences. For the dataset with full exponent range, the current and new values showed major differences (greater than 32 bits) in 567 cases out of 100,000 (0.56%). In every one of these cases, the ratio of c/d or d/c (as appropriate) was zero or subnormal, indicating the advantage of the new method and its continued correctness where needed. PERFORMANCE Test results In order for a library change to be practical, it is necessary to show the slowdown is tolerable. The slowdowns observed are much less than would be seen by (for example) switching from hardware double precison to a software quad precision, which on the tested machines causes a slowdown of around 100x). The actual slowdown depends on the machine architecture. It also depends on the nature of the input data. If underflow/overflow is rare, then implementations that have strong branch prediction will only slowdown by a few cycles. If underflow/overflow is common, then the branch predictors will be less accurate and the cost will be higher. Results from two machines are presented as examples of the overhead for the new method. The one labeled x86 is a 5 year old Intel x86 processor and the one labeled aarch64 is a 3 year old arm64 processor. In the following chart, the times are averaged over a one million value data set. All values are scaled to set the time of the current method to be 1.0. Lower values are better. A value of less than 1.0 would be faster than the current method and a value greater than 1.0 would be slower than the current method. ================================================ Moderate set full set x86 aarch64 x86 aarch64 ======== =============== =============== float 0.59 0.79 0.45 0.81 double 1.04 1.24 1.38 1.56 long double 1.13 1.24 1.29 1.25 ================================================ Table 5: Performance Comparisons (ratio new/current) The above tables omit the timing for the 1 divide and 2 multiply comparison with the 2 divide approach. The float results show clear performance improvement due to using the simple method with double precision for intermediate calculations. The double results with the newer method show less overhead for the moderate dataset than for the full dataset. That's because the moderate dataset does not ever take the new branches which protect from under/overflow. The better the branch predictor, the lower the cost for these untaken branches. Both platforms are somewhat dated, with the x86 having a better branch predictor which reduces the cost of the additional branches in the new code. Of course, the relative slowdown may be greater for some architectures, especially those with limited branch prediction combined with a high cost of misprediction. The long double results are fairly consistent in showing the moderate additional cost of the extra branches and calculations for all cases. The observed cost for all precisions is claimed to be tolerable on the grounds that: (a) the cost is worthwhile considering the accuracy improvement shown. (b) most applications will only spend a small fraction of their time calculating complex divide. (c) it is much less than the cost of extended precision (d) users are not forced to use it (as described below) Those users who find this degree of slowdown unsatisfactory may use the gcc switch -fcx-fortran-rules which does not use the library routine, instead inlining Smith's method without the C99 requirement for dealing with NaN results. The proposed patch for libgcc complex divide does not affect the code generated by -fcx-fortran-rules. SUMMARY When input data to complex divide has exponents whose absolute value is less than half of *_MAX_EXP, this patch makes no changes in accuracy and has only a modest effect on performance. When input data contains values outside those ranges, the patch eliminates more than 99.9% of major errors with a tolerable cost in performance. In comparison to Elen Kalda's method, this patch introduces more performance overhead but reduces major errors by a factor of greater than 4000. REFERENCES [1] Nelson H.F. Beebe, "The Mathematical-Function Computation Handbook. Springer International Publishing AG, 2017. [2] Robert L. Smith. Algorithm 116: Complex division. Commun. ACM, 5(8):435, 1962. [3] Michael Baudin and Robert L. Smith. "A robust complex division in Scilab," October 2012, available at http://arxiv.org/abs/1210.4539. [4] Elen Kalda: Complex division improvements in libgcc https://gcc.gnu.org/legacy-ml/gcc-patches/2019-08/msg01629.html 2020-12-08 Patrick McGehearty gcc/c-family/ * c-cppbuiltin.c (c_cpp_builtins): Add supporting macros for new complex divide libgcc/ * libgcc2.c (XMTYPE, XCTYPE, RBIG, RMIN, RMIN2, RMINSCAL, RMAX2): Define. (__divsc3, __divdc3, __divxc3, __divtc3): Improve complex divide. * config/rs6000/_divkc3.c (RBIG, RMIN, RMIN2, RMINSCAL, RMAX2): Define. (__divkc3): Improve complex divide. gcc/testsuite/ * gcc.c-torture/execute/ieee/cdivchkd.c - New test. * gcc.c-torture/execute/ieee/cdivchkf.c - New test. * gcc.c-torture/execute/ieee/cdivchkld.c - New test. --- gcc/c-family/c-cppbuiltin.c | 50 ++++++- .../gcc.c-torture/execute/ieee/cdivchkd.c | 123 ++++++++++++++++ .../gcc.c-torture/execute/ieee/cdivchkf.c | 122 ++++++++++++++++ .../gcc.c-torture/execute/ieee/cdivchkld.c | 123 ++++++++++++++++ libgcc/config/rs6000/_divkc3.c | 105 ++++++++++++-- libgcc/libgcc2.c | 155 +++++++++++++++++++-- 6 files changed, 654 insertions(+), 24 deletions(-) create mode 100644 gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkd.c create mode 100644 gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkf.c create mode 100644 gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkld.c diff --git a/gcc/c-family/c-cppbuiltin.c b/gcc/c-family/c-cppbuiltin.c index 7b7b07d..8d6a15c 100644 --- a/gcc/c-family/c-cppbuiltin.c +++ b/gcc/c-family/c-cppbuiltin.c @@ -1272,8 +1272,9 @@ c_cpp_builtins (cpp_reader *pfile) { scalar_float_mode mode = mode_iter.require (); const char *name = GET_MODE_NAME (mode); + const int name_len = strlen (name); char *macro_name - = (char *) alloca (strlen (name) + = (char *) alloca (name_len + sizeof ("__LIBGCC__MANT_DIG__")); sprintf (macro_name, "__LIBGCC_%s_MANT_DIG__", name); builtin_define_with_int_value (macro_name, @@ -1281,11 +1282,11 @@ c_cpp_builtins (cpp_reader *pfile) if (!targetm.scalar_mode_supported_p (mode) || !targetm.libgcc_floating_mode_supported_p (mode)) continue; - macro_name = (char *) alloca (strlen (name) + macro_name = (char *) alloca (name_len + sizeof ("__LIBGCC_HAS__MODE__")); sprintf (macro_name, "__LIBGCC_HAS_%s_MODE__", name); cpp_define (pfile, macro_name); - macro_name = (char *) alloca (strlen (name) + macro_name = (char *) alloca (name_len + sizeof ("__LIBGCC__FUNC_EXT__")); sprintf (macro_name, "__LIBGCC_%s_FUNC_EXT__", name); char suffix[20] = ""; @@ -1342,11 +1343,52 @@ c_cpp_builtins (cpp_reader *pfile) default: gcc_unreachable (); } - macro_name = (char *) alloca (strlen (name) + macro_name = (char *) alloca (name_len + sizeof ("__LIBGCC__EXCESS_" "PRECISION__")); sprintf (macro_name, "__LIBGCC_%s_EXCESS_PRECISION__", name); builtin_define_with_int_value (macro_name, excess_precision); + + if ((mode == TYPE_MODE (float_type_node)) + || (mode == TYPE_MODE (double_type_node)) + || (mode == TYPE_MODE (long_double_type_node))) + { + char val_name[64]; + char fname[8] = ""; + if (mode == TYPE_MODE (float_type_node)) + memcpy (fname, "FLT", 4); + else if (mode == TYPE_MODE (double_type_node)) + memcpy (fname, "DBL", 4); + else if (mode == TYPE_MODE (long_double_type_node)) + memcpy (fname, "LDBL", 5); + + if ((mode == TYPE_MODE (float_type_node)) + || (mode == TYPE_MODE (double_type_node))) + { + macro_name = (char *) alloca (name_len + + sizeof ("__LIBGCC_EPSILON__" + )); + sprintf (macro_name, "__LIBGCC_%s_EPSILON__", name); + sprintf (val_name, "__%s_EPSILON__", fname); + builtin_define_with_value (macro_name, val_name, 0); + } + + macro_name = (char *) alloca (name_len + + sizeof ("__LIBGCC_MAX__")); + sprintf (macro_name, "__LIBGCC_%s_MAX__", name); + sprintf (val_name, "__%s_MAX__", fname); + builtin_define_with_value (macro_name, val_name, 0); + + macro_name = (char *) alloca (name_len + + sizeof ("__LIBGCC_MIN__")); + sprintf (macro_name, "__LIBGCC_%s_MIN__", name); + sprintf (val_name, "__%s_MIN__", fname); + builtin_define_with_value (macro_name, val_name, 0); + } +#ifdef HAVE_adddf3 + builtin_define_with_int_value ("__LIBGCC_HAVE_HWDBL__", + HAVE_adddf3); +#endif } /* For libgcc crtstuff.c and libgcc2.c. */ diff --git a/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkd.c b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkd.c new file mode 100644 index 0000000..3419734 --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkd.c @@ -0,0 +1,123 @@ +/* + Program to test complex divide for correct results on selected values. + Checking known failure points. +*/ + +extern void abort(); +extern void exit(); + +extern double log2(); + +#define SMALL (0x1.0p-1024) +#define MAXBIT 52 +#define ERRLIM 6 + +/* + Compare c (computed value) with z (expected value). + Return 0 if within allowed range. Return 1 if not. +*/ +int match(_Complex c, _Complex z) +{ + double rz, iz, rc, ic; + double rerr, ierr, rmax; + double biterr; + rz = __real__ z; + iz = __imag__ z; + rc = __real__ c; + ic = __imag__ c; + + if (__builtin_fabs(rz) > SMALL) + { + rerr = __builtin_fabs(rz-rc)/__builtin_fabs(rz); + } + else if (__builtin_fabs(rz) == 0.0) + { + rerr = __builtin_fabs(rc); + } + else + { + rerr = __builtin_fabs(rz-rc)/SMALL; + } + + if (__builtin_fabs(iz) > SMALL) + { + ierr = __builtin_fabs(iz-ic)/__builtin_fabs(iz); + } + else if (__builtin_fabs(iz) == 0.0) + { + ierr = __builtin_fabs(ic); + } + else + { + ierr = __builtin_fabs(iz-ic)/SMALL; + } + rmax = __builtin_fmax(rerr, ierr); + biterr = 0; + if ( rmax != 0.0) + { + biterr = log2(rmax) + MAXBIT + 1; + } + + if (biterr >= ERRLIM) + return 0; + else + return 1; +} + + +int main(int argc, char** argv) +{ + double _Complex a,b,c,z; + double xr[4], xi[4], yr[4], yi[4], zr[4], zi[4]; + double cr, ci; + int i; + int ok = 1; + xr[0] = -0x1.16e7fad79e45ep+651; + xi[0] = -0x1.f7f75b94c6c6ap-860; + yr[0] = -0x1.2f40d8ff7e55ep+245; + yi[0] = -0x0.0000000004ebcp-1022; + zr[0] = 0x1.d6e4b0e282869p+405; + zi[0] = -0x1.e9095e311e706p-900; + + xr[1] = -0x1.21ff587f953d3p-310; + xi[1] = -0x1.5a526dcc59960p+837; + yr[1] = 0x1.b88b8b552eaadp+735; + yi[1] = -0x1.873e2d6544d92p-327; + zr[1] = 0x1.65734a88b2de0p-961; + zi[1] = -0x1.927e85b8b5770p+101; + + xr[2] = 0x1.4612e41aa8080p-846; + xi[2] = -0x0.0000000613e07p-1022; + yr[2] = 0x1.df9cd0d58caafp-820; + yi[2] = -0x1.e47051a9036dbp-584; + zr[2] = 0x1.9b194f3fffa32p-469; + zi[2] = 0x1.58a00ab740a6bp-263; + + xr[3] = 0x1.cb27eece7c585p-355; + xi[3] = 0x0.000000223b8a8p-1022; + yr[3] = -0x1.74e7ed2b9189fp-22; + yi[3] = 0x1.3d80439e9a119p-731; + zr[3] = -0x1.3b35ed806ae5ap-333; + zi[3] = -0x0.05e01bcbfd9f6p-1022; + + + for (i = 0; i < 4; i++) + { + __real__ a = xr[i]; + __imag__ a = xi[i]; + __real__ b = yr[i]; + __imag__ b = yi[i]; + __real__ z = zr[i]; + __imag__ z = zi[i]; + c = a/b; + cr = __real__ c; + ci = __imag__ c; + + if (!match(c,z)){ + ok = 0; + } + } + if (!ok) + abort(); + exit(0); +} diff --git a/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkf.c b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkf.c new file mode 100644 index 0000000..5cc8ee6 --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkf.c @@ -0,0 +1,122 @@ +/* + Program to test complex divide for correct results on selected values. + Checking known failure points. +*/ + +extern void abort(); +extern void exit(); + +extern float log2f(float); + +#define SMALL (0x1.0p-126) +#define MAXBIT 24 +#define ERRLIM 6 + +/* + Compare c (computed value) with z (expected value). + Return 0 if within allowed range. Return 1 if not. +*/ +int match(float _Complex c, float _Complex z) +{ + float rz, iz, rc, ic; + float rerr, ierr, rmax; + float biterr; + rz = __real__ z; + iz = __imag__ z; + rc = __real__ c; + ic = __imag__ c; + + if (__builtin_fabsf(rz) > SMALL) + { + rerr = __builtin_fabsf(rz-rc)/__builtin_fabsf(rz); + } + else if (__builtin_fabsf(rz) == 0.0) + { + rerr = __builtin_fabsf(rc); + } + else + { + rerr = __builtin_fabsf(rz-rc)/SMALL; + } + + if (__builtin_fabsf(iz) > SMALL) + { + ierr = __builtin_fabsf(iz-ic)/__builtin_fabsf(iz); + } + else if (__builtin_fabsf(iz) == 0.0) + { + ierr = __builtin_fabsf(ic); + } + else + { + ierr = __builtin_fabsf(iz-ic)/SMALL; + } + rmax = __builtin_fmaxf(rerr, ierr); + biterr = 0; + if ( rmax != 0.0) + { + biterr = log2f(rmax) + MAXBIT + 1; + } + + if (biterr >= ERRLIM) + return 0; + else + return 1; +} + + +int main(int argc, char** argv) +{ + float _Complex a,b,c,z; + float xr[4], xi[4], yr[4], yi[4], zr[4], zi[4]; + float cr, ci; + int i; + int ok = 1; + xr[0] = 0x1.0b1600p-133; + xi[0] = 0x1.5e1c28p+54; + yr[0] = -0x1.cdec8cp-119; + yi[0] = 0x1.1e72ccp+32; + zr[0] = 0x1.38e502p+22; + zi[0] = -0x1.f89220p-129; + + xr[1] = -0x1.b1bee2p+121; + xi[1] = -0x1.cb403ep-59; + yr[1] = 0x1.480000p-144; + yi[1] = -0x1.c66fc4p+5; + zr[1] = -0x1.60b8cap-34; + zi[1] = -0x1.e8b02ap+115; + + xr[2] = -0x1.3f6e00p-97; + xi[2] = -0x1.c00000p-146; + yr[2] = 0x1.000000p-148; + yi[2] = -0x1.0c4e70p-91; + zr[2] = 0x1.aa50d0p-55; + zi[2] = -0x1.30c746p-6; + + xr[3] = 0x1.000000p-148; + xi[3] = 0x1.f4bc04p-84; + yr[3] = 0x1.00ad74p-20; + yi[3] = 0x1.2ad02ep-85; + zr[3] = 0x1.1102ccp-127; + zi[3] = 0x1.f369a4p-64; + + for (i = 0; i < 4; i++) + { + __real__ a = xr[i]; + __imag__ a = xi[i]; + __real__ b = yr[i]; + __imag__ b = yi[i]; + __real__ z = zr[i]; + __imag__ z = zi[i]; + c = a/b; + cr = __real__ c; + ci = __imag__ c; + + if (!match(c,z)){ + ok = 0; + } + } + if (!ok) + abort(); + exit(0); +} diff --git a/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkld.c b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkld.c new file mode 100644 index 0000000..74046ee --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/ieee/cdivchkld.c @@ -0,0 +1,123 @@ +/* + Program to test complex divide for correct results on selected values. + Checking known failure points. +*/ + +extern void abort(); +extern void exit(); + +extern long double log2l(long double); + +#define SMALL (0x1.0p-16384L) +#define MAXBIT 63 +#define ERRLIM 6 + +/* + Compare c (computed value) with z (expected value). + Return 0 if within allowed range. Return 1 if not. +*/ +int match(_Complex c, _Complex z) +{ + long double rz, iz, rc, ic; + long double rerr, ierr, rmax; + long double biterr; + rz = __real__ z; + iz = __imag__ z; + rc = __real__ c; + ic = __imag__ c; + + if (__builtin_fabsl(rz) > SMALL) + { + rerr = __builtin_fabsl(rz-rc)/__builtin_fabsl(rz); + } + else if (__builtin_fabsl(rz) == 0.0) + { + rerr = __builtin_fabsl(rc); + } + else + { + rerr = __builtin_fabsl(rz-rc)/SMALL; + } + + if (__builtin_fabsl(iz) > SMALL) + { + ierr = __builtin_fabsl(iz-ic)/__builtin_fabsl(iz); + } + else if (__builtin_fabsl(iz) == 0.0) + { + ierr = __builtin_fabsl(ic); + } + else + { + ierr = __builtin_fabsl(iz-ic)/SMALL; + } + rmax = __builtin_fmaxl(rerr, ierr); + biterr = 0; + if ( rmax != 0.0) + { + biterr = log2l(rmax) + MAXBIT + 1; + } + + if (biterr >= ERRLIM) + return 0; + else + return 1; +} + + +int main(int argc, char** argv) +{ + long double _Complex a,b,c,z; + long double xr[4], xi[4], yr[4], yi[4], zr[4], zi[4]; + long double cr, ci; + int i; + int ok = 1; + xr[0] = -0x9.c793985b7d029d90p-8480L; + xi[0] = 0x8.018745ffa61a8fe0p+16329L; + yr[0] = -0xe.d5bee9c523a35ad0p-15599L; + yi[0] = -0xa.8c93c5a4f94128f0p+869L; + zr[0] = -0xc.248bc228e01adcb0p+15456L; + zi[0] = -0x8.89baf6960fac5cf0p-1011L; + + xr[1] = 0xb.68e44bc6d0b91a30p+16026L; + xi[1] = 0xb.ab10f5453e972f30p-14239L; + yr[1] = 0x8.8cbd470705428ff0p-16350L; + yi[1] = -0xa.0c1cbeae4e4b69f0p+347L; + zr[1] = 0xf.76204243c2f28070p-1022L; + zi[1] = 0x9.15b5abcc93d1f920p+15676L; + + xr[2] = -0x9.e8c093a43b546a90p+15983L; + xi[2] = 0xc.95b18274208311e0p-2840L; + yr[2] = -0x8.dedb729b5c1b2ec0p+8L; + yi[2] = 0xa.a49fb81b24738370p-16385L; + zr[2] = 0x8.efccf744dd88c7a0p+15972L; + zi[2] = 0xa.b8f3f7c826b5b780p-421L; + + xr[3] = 0xc.4687f251c0f48bd0p-3940L; + xi[3] = -0xe.a3f2138992d85fa0p+15598L; + yr[3] = 0xe.4b0c25c3d5ebb830p-16344L; + yi[3] = -0xa.6cbf1ba80f7b97a0p+78L; + zr[3] = 0xb.3c2dd11dfdba2670p+15517L; + zi[3] = -0xf.6773dc63dee9b110p-905L; + + + for (i = 0; i < 4; i++) + { + __real__ a = xr[i]; + __imag__ a = xi[i]; + __real__ b = yr[i]; + __imag__ b = yi[i]; + __real__ z = zr[i]; + __imag__ z = zi[i]; + c = a/b; + cr = __real__ c; + ci = __imag__ c; + + if (!match(c,z)){ + ok = 0; + } + } + if (!ok) + abort(); + exit(0); +} diff --git a/libgcc/config/rs6000/_divkc3.c b/libgcc/config/rs6000/_divkc3.c index 91482af..34563c1 100644 --- a/libgcc/config/rs6000/_divkc3.c +++ b/libgcc/config/rs6000/_divkc3.c @@ -37,31 +37,118 @@ see the files COPYING3 and COPYING.RUNTIME respectively. If not, see #define __divkc3 __divkc3_sw #endif +# define RBIG ((__LIBGCC_TF_MAX__)/2) +# define RMIN (__LIBGCC_TF_MIN__) +# define RMIN2 (__LIBGCC_DF_EPSILON__) +# define RMINSCAL (1/__LIBGCC_DF_EPSILON__) +# define RMAX2 ((RBIG)*(RMIN2)) + + TCtype __divkc3 (TFtype a, TFtype b, TFtype c, TFtype d) { TFtype denom, ratio, x, y; TCtype res; - /* ??? We can get better behavior from logarithmic scaling instead of - the division. But that would mean starting to link libgcc against - libm. We could implement something akin to ldexp/frexp as gcc builtins - fairly easily... */ - if (FABS (c) < FABS (d)) + /* double, extended, long double have significant potential + underflow/overflow errors than can be greatly reduced with + a limited number of tests and adjustments. float is handled + the same way when no HW double is available. + */ + + /* Scale by max(c,d) to reduce chances of denominator overflowing. */ + if (FABS(c) < FABS(d)) { + /* Prevent underflow when denominator is near max representable. */ + if (FABS(d) >= RBIG) + { + a = a / 2; + b = b / 2; + c = c / 2; + d = d / 2; + } + /* Avoid overflow/underflow issues when c and d are small. + Scaling up helps avoid some underflows. + No new overflow possible since c&d < RMIN2. */ + if (FABS(d) < RMIN2) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + else + { + if (((FABS(a) < RMIN) && (FABS(b) < RMAX2) && (FABS(d) < RMAX2)) + || ((FABS(b) < RMIN) && (FABS(a) < RMAX2) && (FABS(d) < RMAX2))) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + } ratio = c / d; denom = (c * ratio) + d; - x = ((a * ratio) + b) / denom; - y = ((b * ratio) - a) / denom; + /* Choose alternate order of computation if ratio is subnormal. */ + if (FABS(ratio) > RMIN) + { + x = ((a * ratio) + b) / denom; + y = ((b * ratio) - a) / denom; + } + else + { + x = ((c * (a / d)) + b) / denom; + y = ((c * (b / d)) - a) / denom; + } } else { + /* Prevent underflow when denominator is near max representable. */ + if (FABS(c) >= RBIG) + { + a = a / 2; + b = b / 2; + c = c / 2; + d = d / 2; + } + /* Avoid overflow/underflow issues when both c and d are small. + Scaling up helps avoid some underflows. + No new overflow possible since both c&d are less than RMIN2. */ + if (FABS(c) < RMIN2) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + else + { + if (((FABS(a) < RMIN) && (FABS(b) < RMAX2) && (FABS(c) < RMAX2)) + || ((FABS(b) < RMIN) && (FABS(a) < RMAX2) && (FABS(c) < RMAX2))) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + } ratio = d / c; denom = (d * ratio) + c; - x = ((b * ratio) + a) / denom; - y = (b - (a * ratio)) / denom; + /* Choose alternate order of computation if ratio is subnormal. */ + if (FABS(ratio) > RMIN) + { + x = ((b * ratio) + a) / denom; + y = (b - (a * ratio)) / denom; + } + else + { + x = (a + (d * (b / c))) / denom; + y = (b - (d * (a / c))) / denom; + } } + /* Recover infinities and zeros that computed as NaN+iNaN; the only cases are nonzero/zero, infinite/finite, and finite/infinite. */ if (isnan (x) && isnan (y)) diff --git a/libgcc/libgcc2.c b/libgcc/libgcc2.c index 61de3b4..4af81bd 100644 --- a/libgcc/libgcc2.c +++ b/libgcc/libgcc2.c @@ -1860,33 +1860,62 @@ NAME (TYPE x, int m) #if defined(L_mulhc3) || defined(L_divhc3) # define MTYPE HFtype # define CTYPE HCtype +# define XMTYPE SFtype +# define XCTYPE SCtype # define MODE hc # define CEXT __LIBGCC_HF_FUNC_EXT__ # define NOTRUNC (!__LIBGCC_HF_EXCESS_PRECISION__) #elif defined(L_mulsc3) || defined(L_divsc3) # define MTYPE SFtype # define CTYPE SCtype +# define XMTYPE DFtype +# define XCTYPE DCtype # define MODE sc # define CEXT __LIBGCC_SF_FUNC_EXT__ # define NOTRUNC (!__LIBGCC_SF_EXCESS_PRECISION__) +# define RBIG ((__LIBGCC_SF_MAX__)/2) +# define RMIN (__LIBGCC_SF_MIN__) +# define RMIN2 (__LIBGCC_SF_EPSILON__) +# define RMINSCAL (1/__LIBGCC_SF_EPSILON__) +# define RMAX2 ((RBIG)*(RMIN2)) #elif defined(L_muldc3) || defined(L_divdc3) # define MTYPE DFtype # define CTYPE DCtype # define MODE dc # define CEXT __LIBGCC_DF_FUNC_EXT__ # define NOTRUNC (!__LIBGCC_DF_EXCESS_PRECISION__) +# define RBIG ((__LIBGCC_DF_MAX__)/2) +# define RMIN (__LIBGCC_DF_MIN__) +# define RMIN2 (__LIBGCC_DF_EPSILON__) +# define RMINSCAL (1/__LIBGCC_DF_EPSILON__) +# define RMAX2 ((RBIG)*(RMIN2)) #elif defined(L_mulxc3) || defined(L_divxc3) # define MTYPE XFtype # define CTYPE XCtype # define MODE xc # define CEXT __LIBGCC_XF_FUNC_EXT__ # define NOTRUNC (!__LIBGCC_XF_EXCESS_PRECISION__) +# define RBIG ((__LIBGCC_XF_MAX__)/2) +# define RMIN (__LIBGCC_XF_MIN__) +# define RMIN2 (__LIBGCC_DF_EPSILON__) +# define RMINSCAL (1/__LIBGCC_DF_EPSILON__) +# define RMAX2 ((RBIG)*(RMIN2)) #elif defined(L_multc3) || defined(L_divtc3) # define MTYPE TFtype # define CTYPE TCtype # define MODE tc # define CEXT __LIBGCC_TF_FUNC_EXT__ # define NOTRUNC (!__LIBGCC_TF_EXCESS_PRECISION__) +#if defined(__LIBGCC_TF_MIN__) +# define RBIG ((__LIBGCC_TF_MAX__)/2) +# define RMIN (__LIBGCC_TF_MIN__) +#else +# define RBIG ((__LIBGCC_XF_MAX__)/2) +# define RMIN (__LIBGCC_XF_MIN__) +#endif +# define RMIN2 (__LIBGCC_DF_EPSILON__) +# define RMINSCAL (1/__LIBGCC_DF_EPSILON__) +# define RMAX2 ((RBIG)*(RMIN2)) #else # error #endif @@ -1994,30 +2023,134 @@ CONCAT3(__mul,MODE,3) (MTYPE a, MTYPE b, MTYPE c, MTYPE d) CTYPE CONCAT3(__div,MODE,3) (MTYPE a, MTYPE b, MTYPE c, MTYPE d) { +#if defined(L_divhc3) \ + || (defined(L_divsc3) && defined(__LIBGCC_HAVE_HWDBL__) ) + + /* Half precision is handled with float precision. + float is handled with double precision when double precision + hardware is available. + Due to the additional precision, the simple complex divide + method (without Smith's method) is sufficient to get accurate + answers and runs slightly faster than Smith's method. */ + + XMTYPE aa, bb, cc, dd; + XMTYPE denom; + MTYPE x, y; + CTYPE res; + aa = a; + bb = b; + cc = c; + dd = d; + + denom = (cc * cc) + (dd * dd); + x = ((aa * cc) + (bb * dd)) / denom; + y = ((bb * cc) - (aa * dd)) / denom; + +#else MTYPE denom, ratio, x, y; CTYPE res; - /* ??? We can get better behavior from logarithmic scaling instead of - the division. But that would mean starting to link libgcc against - libm. We could implement something akin to ldexp/frexp as gcc builtins - fairly easily... */ - if (FABS (c) < FABS (d)) + /* double, extended, long double have significant potential + underflow/overflow errors than can be greatly reduced with + a limited number of tests and adjustments. float is handled + the same way when no HW double is available. + */ + + /* Scale by max(c,d) to reduce chances of denominator overflowing. */ + if (FABS(c) < FABS(d)) { + /* Prevent underflow when denominator is near max representable. */ + if (FABS(d) >= RBIG) + { + a = a / 2; + b = b / 2; + c = c / 2; + d = d / 2; + } + /* Avoid overflow/underflow issues when c and d are small. + Scaling up helps avoid some underflows. + No new overflow possible since c&d < RMIN2. */ + if (FABS(d) < RMIN2) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + else + { + if (((FABS(a) < RMIN) && (FABS(b) < RMAX2) && (FABS(d) < RMAX2)) + || ((FABS(b) < RMIN) && (FABS(a) < RMAX2) && (FABS(d) < RMAX2))) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + } ratio = c / d; denom = (c * ratio) + d; - x = ((a * ratio) + b) / denom; - y = ((b * ratio) - a) / denom; + /* Choose alternate order of computation if ratio is subnormal. */ + if (FABS(ratio) > RMIN) + { + x = ((a * ratio) + b) / denom; + y = ((b * ratio) - a) / denom; + } + else + { + x = ((c * (a / d)) + b) / denom; + y = ((c * (b / d)) - a) / denom; + } } else { + /* Prevent underflow when denominator is near max representable. */ + if (FABS(c) >= RBIG) + { + a = a / 2; + b = b / 2; + c = c / 2; + d = d / 2; + } + /* Avoid overflow/underflow issues when both c and d are small. + Scaling up helps avoid some underflows. + No new overflow possible since both c&d are less than RMIN2. */ + if (FABS(c) < RMIN2) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + else + { + if (((FABS(a) < RMIN) && (FABS(b) < RMAX2) && (FABS(c) < RMAX2)) + || ((FABS(b) < RMIN) && (FABS(a) < RMAX2) && (FABS(c) < RMAX2))) + { + a = a * RMINSCAL; + b = b * RMINSCAL; + c = c * RMINSCAL; + d = d * RMINSCAL; + } + } ratio = d / c; denom = (d * ratio) + c; - x = ((b * ratio) + a) / denom; - y = (b - (a * ratio)) / denom; + /* Choose alternate order of computation if ratio is subnormal. */ + if (FABS(ratio) > RMIN) + { + x = ((b * ratio) + a) / denom; + y = (b - (a * ratio)) / denom; + } + else + { + x = (a + (d * (b / c))) / denom; + y = (b - (d * (a / c))) / denom; + } } +#endif - /* Recover infinities and zeros that computed as NaN+iNaN; the only cases - are nonzero/zero, infinite/finite, and finite/infinite. */ + /* Recover infinities and zeros that computed as NaN+iNaN; the only + cases are nonzero/zero, infinite/finite, and finite/infinite. */ if (isnan (x) && isnan (y)) { if (c == 0.0 && d == 0.0 && (!isnan (a) || !isnan (b)))